Unicode Truth Engine
Your filesystem lies, your version control does not notice, and your cloud silently mutates names. UTE restores round-trip-safe filename identity across every platform.
Round-trip
Identity preserved across platforms
NFC · NFKC
Normalization tracked, not guessed
Homoglyph
Spoofing made observable
Every platform reads a filename differently.
Unicode defines characters and normalization forms but not how platforms apply them. HFS+ forces NFD, NTFS and ext leave raw bytes, S3 treats keys opaquely, and Git has no Unicode model at all. The same name diverges as it crosses systems, so visibility, comparison and identity quietly break.
The gap is also a security surface. Homoglyphs — Cyrillic а against Latin a — drive supply-chain typosquatting, and a sort-versus-lookup mismatch (Unicode collation at scan time, byte comparison at runtime) is a TOCTOU class that hides shadow files from a scanner but still runs them — the same mismatch Apple had to fix by abandoning binary search over normalized filenames.
Preserve intent, explicitly.
01
A typed transformation pipeline
UTE captures the raw syscall bytes, normalizes to a tagged internal form, persists a canonical on-disk form, and derives a separate collation form for rendering — each step type-enforced and reversible, so intent is never lost between platforms.
02
Fingerprint the standard itself
The Unicode Character Database drifts with every release. UTE records the UCD version used at each transform, so a name stays reversible even as the standard moves under it.
03
Make spoofing observable
A security-audit mode diffs homoglyph sets by script origin and aligns sort order with access semantics — turning an invisible attack surface into something you can see, warn on, and block.
Research, stated as research.
| Milestone | Status | Evidence |
|---|---|---|
| Typed transform pipeline | In progress | Internal |
| Filesystem coverage (APFS/NTFS/ext/S3) | In progress | Divergence mapped |
| Security-audit mode (homoglyph / TOCTOU) | In progress | Prototype |
| Git + object-store integration | Next | — |
Figures on this page are targets and internal measurements at proof-of-concept stage. We share the methodology with anyone who wants to run it.
Where names cross boundaries.
Next is canonical diff tooling and integrations for the two worst offenders — Git and object stores — plus forensic visualization of how a name mutates as it travels.
Unicode Truth Engine spins out of Research ByIQ once round-trip fidelity holds across the major filesystems and cloud stores.